1. Field of the Invention
The present invention relates to an optical disc apparatus for recording information by forming on a recording medium a recording mark with a different physical property from that of other parts of the medium.
2. Description of Related Art
With increases in the speed and density of optical discs, the PRML (Partial Response Maximum Likelihood) read signal processing method is becoming more and more indispensable. As an example of the PRML method, an adaptive or compensated PRML method is known whereby a target signal level is adaptively varied in accordance with the read signal. Tech. Digest ISOM '03, pp. 34, indicates that, by compensating the asymmetry in the read signal and the thermal interference during recording by such a PRML method, the capacity of a Blu-ray Disc-compatible apparatus can be increased by approximately 35 GB.
In optical disc apparatuses utilizing a PRML device as well as in conventional apparatuses, it is important to cause the apparatus to appropriately study: (1) reproduction equalization conditions; (2) focus position and tilt conditions; and (3) write power and pulse conditions, for example. In this case, it is necessary to optimize various parameters using an evaluation index for read signal quality such that the aforementioned conditions can be optimized. Conventionally, when the direct-slice reading method is used, jitter has been used as a read signal quality evaluation index. In Tech. Digest ODS '03, pp. 93 and in Tech. Digest ISOM '03, pp. 116, as the evaluation index of the quality of read signal that is adapted to the PRML method, MLSE (Maximum Likelihood Sequence Error) is indicated corresponding to the PR(1,2,2,1)ML channel. Using a correct bit sequence as the result of decoding and an error bit sequence obtained by shifting the correct bit sequence by one bit, the Euclidean distance between the read signal and each sequence is calculated to evaluate the read signal from the view point of probability of error. It is also indicated that, since MLSE focuses on data edge, a recording strategy that utilizes adaptive write pulse conditions that are tabulated depending on the preceding and subsequent space and mark lengths can be optimized by measuring and evaluating the shift of the MLSE value from a target value for each of the elements in the table. Furthermore, in Tech. Digest ISOM '03, pp. 164, PRSNR (Partial Response Signal to Noise Ratio) is indicated for the PR(1,2,2,2,1)ML channel. This is a technique whereby, in the PR(1,2,2,2,1) channel, three patterns with small Euclidean distances and high error frequencies are extracted, and the value of each Euclidean distance is calculated, thereby calculating a read signal SNR from the viewpoint of probability of error and evaluating signal quality. The publication indicates that there is an excellent correlation between PRSNR and bit error rate.
Hereafter, in order to help better understand the invention, the evaluation of error detection in the PRML method is described.
FIG. 2 shows some of the bit error patterns in the case where the RLL(1,7) code is decoded using the PR(1,2,2,1) class. In this case, since the number of bits representing the class (to be hereafter referred to as class bit number N) is four, 7-bit (2N-1) patterns may be considered to take the influence of one-bit error into account. A pattern with a different center bit from that of a correct pattern is referred to as an error pattern. When conditions where the correct pattern and error pattern each satisfy the run-length limit are extracted, there are eight combinations of the patterns with respect to a one-bit error, as shown in FIG. 2. The value obtained by adding up the square of the difference between the target signal levels of the correct and error patterns at each time is referred to as the Euclidean distance between them, which would be (12+22+22+12=14) or 14 in the case of the one-bit error. When normalized such that the amplitude of the target signal becomes 2, the Euclidean distance would be 1.11. If the temporal transition of target values of the two bit patterns are thought of as vectors of M dimensions (M=4 in the present case), the Euclidean distance could also be regarded as the distance between two points in a space including the aforementioned vectors as position vectors. For a 2-bit error, there would be 12 combinations, and the Euclidean distance would be 14. Similarly, if increasingly more complicated error patterns are considered, the relevant Euclidean distance would continue ad infinitum from 16, 18, 20, 22, and so on. Statistically, errors of all these patterns would be produced. However, a huge volume of processing would be required to evaluate signal quality if all these error patterns are to be included, and it would be impossible to implement all the required processes on an optical disc drive. Since the Euclidean distance is the distance between a correct pattern and an error pattern, it may be considered to be an index of the unlikelihood of presence of error. In fact, within a range in which error correction is possible, such as in a range of bit error rate of not more than about 10−4, errors in patterns with the minimum Euclidean distance are dominant. Thus, it is fair to say that sufficient evaluation of signal quality can be made by evaluating only the patterns with the minimum Euclidean distance. MLSE only focuses on the minimum Euclidean distance pattern, namely a 1T edge shift in the case of PR(1,2,2,1) and measures the distribution of the likelihood of presence of error on a location by location basis so that the standard deviation of the distribution can be evaluated by approximation with respect to a normal distribution.
Similarly, FIGS. 3 and 4 each show a summary of error patterns and Euclidean distance for PR(1,2,1) and PR(1,2,2,2,1), respectively, with respect to the RLL(1,7) code.
FIG. 3 shows a summary of the error patterns and associated Euclidean distances for PR(1,2,1). As shown, the pattern of the minimum Euclidean distance is similarly a 1T edge shift, and the Euclidean distance is 6. This can also be evaluated by MLSE, as in PR(1,2,2,1).
FIG. 4 shows a summary of the error patterns and Euclidean distance for PR(1,2,2,2,1). As shown, whereas the Euclidean distance for 1-bit error is 14, the Euclidean distance for 2-bit and 3-bit error patterns is 12. In this case, it can be considered that MLSE, which evaluates only 1T shift errors, is incapable of accurately evaluating signal quality. Thus, in Tech. Digest ISOM '03, pp. 164, the difficulty with which error can occur in these three patterns is quantified from the viewpoint of S/N, and signal quality is evaluated using one of the patterns that has the smallest S/N and that is likely to result in error. This is how the aforementioned PRSNR is utilized.
FIG. 5 shows some of the bit error patterns in the case where the RLL(1,7) code is decoded using the PR(1,1,1,1) class, which is generally used. In this case, since the class bit number N is 4, 7-bit patterns may be considered in order to take the influence of a 1-bit error into account, as in the case of the PR(1,2,2,1) class. As shown, there are 8 combinations of the patterns with respect to a 1-bit error, and the Euclidean distance is 4. There are 18 combinations of the case where a 2-bit error is produced with respect to 10-bit patterns, and the Euclidean distance is 4. Similarly, if increasingly more complicated patterns are considered, the corresponding Euclidean distance would increase ad infinitum, from 6, 10, . . . and so on. In this case, it is necessary to consider patterns with the minimum Euclidean distance for not only 1-bit errors but also 2-bit errors.    Non-Patent Document 1: Tech. Digest ISOM '03, pp.34    Non-Patent Document 2: Tech. Digest ODS '03, pp. 93    Non-Patent Document 3: Tech. Digest ISOM '03, pp. 116    Non-Patent Document 4: Tech. Digest ISOM '03, pp. 164